High-Resolution 3D Reconstruction
We developed a binocular stereo method which is optimized for reconstructing surface detail and exploits the high image resolutions of current digital cameras. Our method occupies a middle ground between stereo algorithms focused at depth layering of cluttered scenes and multi-view ”object reconstruction” approaches which require a higher view count. It is based on global non-linear optimization of continuous scene depth rather than discrete pixel disparities. We use a mesh-based data-term for large images, and a smoothness term using robust error norms to allow detailed surface geometry. The continuous optimization approach enables interesting extensions beyond the core algorithm: Firstly, with small changes to the data-term camera parameters instead of depth can be optimized in the same framework. Secondly, our approach is well suited for a semi-interactive reconstruction work-flow, for which we propose several tools.
Left to right: (1) Complete set of input images, 12 megapixels, shot freehand in a museum under available light; (2) depth map representation of our reconstruction; (3) reconstruction rendered as shaded mesh; (4,5) detail crop of the left view and the shaded reconstruction.
Initialization from sparse interest point correspondences (top left) and convergence over 20 iterations. Many surface details appear in the first iteration (top right).
Results computed with the proposed method, shown as shaded meshes rather than depth maps to emphasize surface detail.
Videos
Publications
D. Blumenthal-Barby, P. Eisert
High-Resolution Depth For Binocular Image-Based Modelling, Computers & Graphics , vol. 39, pp. 89-100, Apr. 2014. [PDF]